AMD Threadripper 2990WX 32-Core and 2950X 16-Core Review • Hacker News Board

Re: AMD Threadripper 2990WX 32-Core and 2950X 16-Core Review

#81

Post by ksec » Mon, Aug 13, 2018, 3:43 PM UTC

srcmap wrote: post #40

Two additional benchmarks I like to see from Anandtech: 1) some VM base tests for these kind of CPU - 32, 64, 128 VM all running some kind of web/db/redis benchmarks inside. 2) Some compilation testing - time the clean build of AOSP, BSD, some very complex linux app - use max jobs setting for parallel compile and time how long it take to finish the jobs. ( measure the over CPU usages at the same time. )

For these kind of test I think Servethehome.com and https://www.phoronix.com does a better job.

Re: AMD Threadripper 2990WX 32-Core and 2950X 16-Core Review

#82

Post by pokemongoaway » Mon, Aug 13, 2018, 3:43 PM UTC

jdietrich wrote: post #26

karangoeluw wrote: post #2

Honest q: What home-use workstation would use 32 cores? (Excluding home labs or servers).

Digital audio workstation workloads are massively multithreaded, with hundreds or thousands of DSP processes. Performance scales almost linearly with core count.

And if only a great DAW ran on Linux! We'd have lots of competitive optimization options...

Re: AMD Threadripper 2990WX 32-Core and 2950X 16-Core Review

#83

Post by ksec » Mon, Aug 13, 2018, 3:45 PM UTC

blattimwind wrote: post #12

> After core counts, the next battle will be on the interconnect. Low power, scalable, and high performance: process node scaling will mean nothing if the interconnect becomes 90% of the total chip power. I think this means that scaling processors to more cores (nodes) in a lightweight-NUMA scheme just isn't going to work. My guess going forward is that the core counts will stay at the current level for quite a while…

It will also means many consumer software can now optimise from low core count to a 64 Core / 128 Thread scenario. And this is likely be a decade worth of work.

Re: AMD Threadripper 2990WX 32-Core and 2950X 16-Core Review

#84

Post by ksec » Mon, Aug 13, 2018, 3:48 PM UTC

Filligree wrote: post #13

The biggest benefit I've found to running a 1950X isn't something I would have expected, but which perhaps I should have. True, it's much faster than my old system for batch processing, but most of the time it's still idle. Even if it's running Chrome flat-out, as far as the 1950X is concerned, that's idle. Because the two NUMA nodes are ~entirely independent, it's capable of running two independent processes at full…

> Folklore would have it that single-thread performance is the most important aspect of desktop performance, but that isn't what I've observed.

It still is. It is only that we have approach the end of the performance limit, or it has approached the same as its competitor that you want to trade double the core count vs ~20% more maximum single thread performance.

Re: AMD Threadripper 2990WX 32-Core and 2950X 16-Core Review

#85

Post by gok » Mon, Aug 13, 2018, 3:50 PM UTC

The Chromium compile being faster on 16 cores than 32 cores is pretty weird, given how embarrassingly parallelizable that should be. Wonder if it's out of memory, or if the bottleneck is actually linking.

Re: AMD Threadripper 2990WX 32-Core and 2950X 16-Core Review

#86

Post by jdietrich » Mon, Aug 13, 2018, 3:51 PM UTC

deng wrote: post #69

jdietrich wrote: post #26

Earlier quoted context omitted.

Digital audio workstation workloads are massively multithreaded, with hundreds or thousands of DSP processes. Performance scales almost linearly with core count.

Not my experience. I've had stutters with Ableton Live on a 4 core machine with 2 cores still idling around. Ableton cannot multi-thread a single track (at least in Version 9), and if you're using a single-threaded VST that does not matter anyway.

If a single instance of a VST can use 100% of a thread, you're using a woefully underpowered processor or a ludicrously inefficient plugin. Many composers regularly work on projects with hundreds of tracks and thousands of plugin instances. Projects of that scale used to require multiple computers and a bunch of DSP accelerator cards, but they're now entirely feasible on one high-end workstation.

Re: AMD Threadripper 2990WX 32-Core and 2950X 16-Core Review

#87

Post by mastax » Mon, Aug 13, 2018, 3:52 PM UTC

blattimwind wrote: post #12

> After core counts, the next battle will be on the interconnect. Low power, scalable, and high performance: process node scaling will mean nothing if the interconnect becomes 90% of the total chip power. I think this means that scaling processors to more cores (nodes) in a lightweight-NUMA scheme just isn't going to work. My guess going forward is that the core counts will stay at the current level for quite a while…

I was amazed at how much power the interconnect was taking on these parts. Even with all the other downsides I was convinced it was the right move, but those power requirements are giving me second (third?) thoughts.

The 7nm parts (incl. Ryzen on up) are rumored to use 16-core CCXes, so we should see 64-core EPYC and possibly Threadripper with no worse IC power use. Hopefully they'll make the IC better as well.

7nm EPYC being fabbed on TSMC does give some chance for a big monolithic die. TSMC has a long history making large dies and their 7nm process should be mature enough for decent yields at that point. I am speculating out of my depth, though.

Re: AMD Threadripper 2990WX 32-Core and 2950X 16-Core Review

#88

Post by xvf22 » Mon, Aug 13, 2018, 3:56 PM UTC

It seems that they left the plastic cover for at least some of the benchmarks [1]. I can imagine that would limit performance since the CPU would be throttling itself down to keep cool.

[1] https://www.anandtech.com/comments/13124/the-amd-threadrippe...

Re: AMD Threadripper 2990WX 32-Core and 2950X 16-Core Review

#89

Post by mrkgnao » Mon, Aug 13, 2018, 3:59 PM UTC

Filligree wrote: post #13

The biggest benefit I've found to running a 1950X isn't something I would have expected, but which perhaps I should have. True, it's much faster than my old system for batch processing, but most of the time it's still idle. Even if it's running Chrome flat-out, as far as the 1950X is concerned, that's idle. Because the two NUMA nodes are ~entirely independent, it's capable of running two independent processes at full…

> Because the two NUMA nodes are ~entirely independent, it's capable of running two independent processes at full speed.

I don't understand. From my (admittedly little better than layperson's) knowledge, I'm guessing the cores of most multicore processors have to compete for memory access...? Is there a good search term I can use to help me understand what's going on here?

Re: AMD Threadripper 2990WX 32-Core and 2950X 16-Core Review

#90

Post by compilerdev » Mon, Aug 13, 2018, 4:00 PM UTC

gok wrote: post #85

The Chromium compile being faster on 16 cores than 32 cores is pretty weird, given how embarrassingly parallelizable that should be. Wonder if it's out of memory, or if the bottleneck is actually linking.

The Chrome compilation looks really strange indeed, especially with having the results from Phoronix on Linux which show that it's the fastest compiling the Linux kernel. I wonder what kind of Chrome build it's doing - does it use Clang-cl or Visual C++? Does it have LTO (LTCG for VC++) enabled? If it's VC++ with LTCG, for example, the entire code generation and linking is limited to 4 cores by default.